[R] Negative binomial theta

[Prof Brian Ripley]

On Tue, 12 Aug 2003, Martin Hoyle wrote:

> I'm trying to use the command "glm.nb" in library(MASS) to test for a
> significant difference in the aggregation parameter "theta" between the
> three levels of a factor.

You cannot in general do that: each fit has one theta parameter.

You may be able to do three separate fits, one for each level, and sum the 
loglikelihoods and do a likelihood ratio test by hand.  (It depends 
exactly what the model you are using is.)


-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595